Resolve 21 IntelliSearch vs a transcript with word times

DaVinci Resolve 21 lists IntelliSearch as a new way to find content in your media. Sume has no search index of your library. What it does have is video-inspect with transcribe, which returns a transcript with text, words[] and optional sentence segments[]. Search that text yourself, take the word time, and pass it to the trim route as start.

From a phrase to a cut

The inspect call returns a probe, stills as durable image artifacts, and, when you ask, a transcript with an audio_url. The probe-only call is free; transcription is billed per minute, so confirm the rate in GET /v1/catalog. Once you have words[], find the first word of the phrase you want and use its start time. The trim route needs Sume-hosted media, start, and exactly one of end or duration, plus an Idempotency-Key.

What you do not get

IntelliSearch is described as a search feature for content in the app. Sume returns evidence, not a search box: you get eight mid-bin stills by default and a transcript on request. Searching for a visual object without speech means reading the stills yourself or sampling more with fps up to 2, capped at 24 frames.

Honest limits

• Blackmagic's page does not say what IntelliSearch indexes, so no equivalence is claimed.
• A transcript only helps when the clip has speech; silent footage returns no words.
• Source clips for inspect and frames are limited to 300 seconds in the docs.